Is AI ruining our skills? Early results are in—and they’re not good

AI-driven deskilling – New survey data and two small-but-signal-loud studies suggest that relying on AI can make people less capable when the tools aren’t there—raising urgent questions about what happens to hard-won expertise in medicine and computer science.

For some medical specialists and software engineers, the fear doesn’t start with a bad outcome. It starts with a simple scenario: what if the system isn’t available?

That worry is already widespread in health care. A survey of US health-care workers published earlier this month found that 70% of nurses and 77% of physicians are worried about losing their skills because of over-reliance on AI systems.

In medicine, the concern is no longer just hypothetical. A study of physicians in Poland who specialize in endoscopy—using flexible probes to examine the inside of the human body—looked at how quickly AI assistance can change performance. All of the physicians had performed at least 2,000 colonoscopies during their careers.

They were given access to an AI system that analyzes colonoscopy images in real time and flags a type of precancerous intestinal lesion called an adenoma. Crucially, the tool wasn’t always available. It was offered to the specialists on some days but not on others.

When physicians began using the system, their performance dropped significantly whenever it was unavailable. During the three-month period before the AI tool was introduced, the specialists found at least one adenoma during 28.4% of colonoscopies. In the three-month period after the tool was introduced. the adenoma detection rate for colonoscopies performed without AI assistance decreased to 22.4%.

The findings, published last October in The Lancet Gastroenterology and Hepatology, point to a troubling pattern: even highly skilled clinicians may get worse at the tasks their jobs require as dependence on AI grows.

That effect comes with a suggested mechanism. The study authors say continuous exposure to such tools can lead clinicians to become “less motivated, less focused, and less responsible when making cognitive decisions without AI assistance”.

Robert Wachter, a physician at the University of California, San Francisco, author of a book on how AI tools are transforming health care, frames the implication in terms of learning and retention, arguing that the study suggests skills can erode as reliance increases.

Yuichi Mori. a physician-researcher at the University of Oslo and a co-author on the paper. cautions that more studies are needed to confirm the phenomenon. But he also warns users to be aware of the risk. “There is no established solution against deskilling right now. It should be a very hot research topic in the next decade.”.

The same anxiety shows up in computer science, where the stakes are different but the logic is similar: if AI helps you complete tasks, will you still learn the underlying concepts?

To probe that. researchers at the AI firm Anthropic in San Francisco. California. designed a randomized controlled trial with 52 software engineers. The engineers were asked to perform a basic coding task. During the exercise, all 52 participants could search the web and access instructions on how to do the task. Half of the participants were also prompted to use an AI assistant.

Afterwards, all participants took a quiz about what they had learnt from the task. Those who had used an AI assistant did significantly worse on the quiz than those who hadn’t: the average score was 50% in the AI group versus 67% in the non-AI group. The AI-assisted participants performed particularly poorly on questions that required them to diagnose errors in the code. suggesting they had failed to learn the concepts behind the code they had just produced.

The study was posted on the preprint server arXiv ahead of peer review.

For Kevin Crowston. an information scientist at Syracuse University in New York and a researcher into how generative AI tools change the way software developers learn and retain coding skills. the results land as a specific kind of unease. He puts it bluntly: “Now you have this very odd disconnect between performance and learning. People can perform at a pretty high level. because they’re basically borrowing skills from the AI. but they are not developing those skills themselves.”.

The wider concern isn’t limited to medicine or coding. Other technologies have reshaped skill sets before—GPS navigation systems, for instance, have eroded people’s navigation skills. Tapani Rinta-Kahila. an information-systems researcher at the Hanken School of Economics in Helsinki. argues that generative AI tools are different because they automate cognitive faculties around thinking and interpretation—capabilities long considered uniquely human.

His own work points in the same direction. In 2018. he published a study of a group of accountants who had been using an automated. non-AI accounting system continuously for more than a decade. When the tool was taken away. his team found that the accountants had forgotten how to do several routine work tasks.

Rinta-Kahila anticipates that AI systems will affect work as they take over basic tasks once performed by early-career professionals. “Next generations of programmers may not understand the foundations of coding that well at all. if they lack the hands-on experience. ” he says. “The same goes for many other knowledge-intensive professions, such as accounting and law.”.

So what can people do right now?. Crowston and Rinta-Kahila both land on a practical message: awareness and restraint. Rinta-Kahila says people need to be aware of how much they are offloading to generative AI tools. They also need to understand how generative AI models work. including their limitations. and avoid trusting AI outputs without questioning them.

He frames the challenge as managing competing dynamics—using tools while staying mentally alert: “People need to manage the competing dynamics of relying on generative AI and staying mindfully vigilant.”

Across these studies and interviews, one question stays hard to ignore: when AI isn’t there—when access ends, or the system fails, or training depends on what you retain on your own—what exactly remains of the skills you thought you were building?

This article is reproduced with permission and was first published on June 18, 2026.

AI deskilling medical specialists adenoma detection rate colonoscopy endoscopy software engineers randomized controlled trial arXiv generative AI learning retention GPS navigation skills